International Migration and Climate Change

Transcription

1 University Jaume I From the SelectedWorks of Inma Martinez-Zarzoso 2012 International Migration and Climate Change Inmaculada Martinez-Zarzoso Available at:

2 International Migration and Climate Change (Preliminary and incomplete) I. Martinez-Zarzoso, University of Goettingen, Germany C. Murris, Department of Economics, Simon Fraser University, Canada A. Backhaus, Toulouse School of Economics, France Abstract This paper uses international migration data and climate variables in a multi-country setting to investigate to what extent migration can be explained with climate variations. A gravity model of migration augmented with average temperature and precipitation in the origin country is estimated including the usual control variables. Several panel data estimators are applied, including dynamic panel data models. We find two primary results. First, higher temperatures increase migration for all countries. Second, lower precipitation levels also increase migration both in the short and in the long term. Both effects are significant in statistical and in economic terms. In particular, a 1 C rise in temperature in a given year increases the migration stock by 3.9 percent for each destination on average. JEL Code: F10 Key Words: temperature precipitation panel data migration flows climate degradation 1. Introduction The nexus between climate change and migration have been addressed since the early 1990s by political scientist, environmentalist and demographers. In particular, the discussion intensified after the release of the conclusions of the Fourth Assessment Report of the IPCC. The report refers to the potential for population migration due to climate distress. The topic has received substantial media coverage but limited academic research. There is in the recent literature a number of policy publications concerning the impacts of climate change on migration, but most of them are case studies for specific regions/countries and time episodes (World Bank, 2010). Surprisingly, the number of empirical studies quantifying this impact is scarce. The standard statistical migration literature examines socioeconomic drivers of bilateral migration without considering climate factors. Only recently, a number of studies focus 1

3 on natural disasters and extreme events as drivers of migration (e.g. Marchiori, Maystadt and Schumacher, 2012; Warner, Stal, Dun and Afifi, 2009). In this paper we focus on permanent migration due to gradual climate change. We follow a similar approach to Dell et al (2008), but our impact variable is migration instead of economic growth. The main aim of the paper is to derive estimates of the effect of changes in climate variables, namely temperature and precipitation on migration flows over time and across countries. In particular, we estimated a gravity-type model extended with climate variables and using the usual control variables and the most recently suggested modeling strategy. To the best of our knowledge this is the first paper to empirically assess the effect of changes in average temperature and average precipitation on international migration. A paper closely related to our work is the gravity study by Afifi and Wagner (2008) that estimates a gravity model augmented with environmental factors, which have in general a significant and positive impact on migration. The rest of the paper is structured as follows. Section 2 revises the academic literature examining the determinants of bilateral migration, which included climate variables. Section 3 refers to the related theoretical models and derives the empirical specification. Section 4 presents the empirical application, describing the data and variables and presenting the main results and Section 5 concludes. 2. Literature review In this section we distinguish between studies focusing on traditional determinants of international migration and new studies adding climate variables. Within the first category, recent studies on migration from many origins to many destinations include Hatton and Williamson (2003) for 21 countries in Africa over the period and 2

4 Mayda (2010) for 79 countries to 14 OECD over the period Warin and Svaton (2008) present and empirical assessment of bilateral migration flows into the EU-15 countries from 76 origin countries during the period More recently, Ruyssen, Everaert, and Rayp (2011) investigate the determinants of bilateral immigrant flows to 19 OECD countries between 1998 and 2007 from both advanced and developing origin countries paying particular attention to dynamics by including both the lagged migrant flow and the migrant stock to capture partial adjustment and network effects. The above mentioned papers ignore weather disasters. Early works relating international migration and natural disasters are overwhelmingly qualitative. There have been only three large statistical studies. Naude (2008) includes 43 African countries in origin to all destinations, but not bilateral emigration and 5-year intervals during the period The main measure used is the sum of weather and seismic disasters in the origin. Reuveny and Moore (2009) cover instead 107 countries to 15 OECD host countries over the period , using bilateral migration data. Non-weather disasters & destination disasters are excluded from this study. However, the short time period, small range of countries, and model limitations make their results difficult to generalize. The most comprehensive study is Alexeev, Good and Reuveny (2010) who cover 21 years ( ) and many origins and destinations. They find that when number of people affected in origin rises by 1 standard deviation (16 million) above its mean, emigration from the origin rises by 23 percent. As regards international migration and climate change, Afifi and Wagner (2008) estimate a gravity model augmented with environmental factors for a cross-section of countries in All the environmental factors considered except for floods have a significant positive impact on migration. In particular, soil quality and availability of suitable water are among the most significant according to the authors. 3

5 3. Theoretical background and model specification A Gravity model for multi-origin and destination countries augmented with climate variables is estimated over the period The basic empirical specification contains as control variables those derived from the neoclassical theory; namely demographic, geographic and cultural controls, as well as the trade share, ln ln ln ln GDPH ln GDPH ln (1) where wtem it denotes Population weighted average annual temperature in degrees C. Constant 1990 population weights are used in constructing the average temperature. wpre it denotes Population weighted average annual precipitation in mm. Note that constant 1990 population weights are used in constructing the average temperature. GDPH it (GDPH jt ) denotes GDP per capita in the origin country (destination country) at year t. Dem_Pre it denotes the share of young people in working age population U it (U jt ) denotes the unemployment rate in origin (destination) at time t. Trade_to_GDP it denotes the openness ration in origin at time t. Ln Dist ij denotes geographical distances between origin and destination countries. Contig ij, Lang ij, Same_Cont ij amd Colony ij are dummy variables that take the value of one when countries i and j share a border, an official language, are in the same continent or have ever had a colonial relationship, respectively, zero otherwise. 4

6 Time ( ) and country-pair ( fixed effects will be included in the estimated model. The latter will preclude us from obtaining estimated coefficient for the bilateral time invariant variables. is the error term that is assumed to be independently and identically distributed. A second specification will be estimated to model time-variant multilateral resistance, as suggested in the trade literature (e.g. Baldwin and Taglioni, 2006). We add to the model destination-and-time dummy variables to be able to control for all the pull factors attached to the host countries that are time variant. As regard the push factors, we leave them in the model since our target variables vary overtime and per origin. The model is specified as, ln ln ln ln GDPH ln (2) where denote destination-and-time dummy variables. Finally a third specification adds dynamics and network effects as suggested in recent literature (Dunlevy, 1993, Ruyssen et al., 2011). We will add as explanatory variables two measures of past immigration. The stock of migrants is considered as a proxy for network effects and the lagged migration flow is specified to capture a partial adjustment mechanism. ln ln, ln, ln ln ln GDPH ln GDPH ln (3) where : 5

7 , Is the stock of migrants from source country i to destination country j at time period t ln, is the past flows of migrants from i to j during the time periods t k, k < t. Cleary, the latter is also contained in the former, as changes in stocks = inflows outflows + births deaths. Therefore we expect to have colinelarity problems in the estimation. 4. Empirical Application 4.1 Data and Variables Average temperature and average precipitation both are from Dell et al (2008). The authors used geospatial software to aggregate both variables to the country-year level. The variables they use are population-weighted averages using 1990 population figures. The data on migration flows from source to destination countries and on stocks in destination countries stems mainly from the OECD s International Migration Database (IMD), in the following referred to as OECD (2010). It covers yearly inflows from 1995 till 2008, yielding 14 time periods for our analysis. However, given that the climate variables are only available until 2006 we restrict our sample to the period We select 19 OECD members as destination countries and examine inflows from 161 countries of origin. Some of the latter are members of the OECD as well; e.g. Poland and Mexico. Although they might themselves be important destination countries from the perspective of less developed countries, their role as sending countries is much more important. A complete list of the source and destination countries can be found in the appendix (Table A.1). The IMD is constructed on the basis of statistical reports of the OECD member countries, which implies that the data might not be fully internationally comparable, as the criteria for registering population and the conditions for granting residence permits vary across countries. The same is true for data on immigrant stocks, as e.g. France counts the stock of immigrants according to their countries of birth, while Germany counts it according to the immigrants nationalities. In addition to these potential 6

8 problems, illegal migration flows are to a great extent only partially covered, as data can only be obtained through censuses. Further, the data exhibits quite serious gaps for some countries. Apparently, the majority of destination countries did not record immigrants from the full set of source countries during the first few years of our period of analysis, as missing data is most frequent in this period. Japan and the Republic of Korea represent two negative extremes, as data is completely missing for most of the destination countries till about 2006 in the case of Japan and until today in the case of the Republic of Korea. Only the inflows from the most important regional destination countries, as e.g. China, the Russian Federation, the Philippines and Indonesia, have been recorded over a longer period of time. But we might still find solace in the presumption that immigration from Africa and South America is generally of much less importance for these countries than it is for Europe and North America. Regarding the European destination countries, data for Austria and Switzerland is mostly limited to European source countries, while Belgium has only recorded inflows from 19 source countries. Data on inflows into Italy is also missing for many source countries and is completely unavailable for the years , 2003 and The case of Portugal is quite puzzling, as data recording appears to have started in 2001 for most source countries. In the same year, Portugal experienced an immigration shock from virtually all source countries according to the data. As only very few inflows had been recorded before, this results in a sharp increase and peak of immigration in 2001 and a similar decrease and normalization at a much lower level in the following years. If this was due to a general measurement error in 2001 or if this reflects the actual immigration patterns remains questionable. Maybe surprising regarding the generally important role as a destination country, the data for migration to the United Kingdom (UK) seems to be very poor. For most source countries, it is only available shortly before and after the year Further, many figures appear to have been rounded. Altogether, these problems might reflect the 7

9 neglecting attitude of several European countries towards exact statistical research on immigration. Fortunately, we were able to fill some of the gaps with observations from the Eurostat online database (referred to as Eurostat (2010) that are for unknown reasons not contained in OECD (2010). For Austria, Switzerland and the UK, numerous non-european source countries could be added. Further, several of the rounded and apparently highly overestimated figures from the UK could be replaced. Adding and replacing rounded observations was only done if the figures from OECD (2010) and Eurostat (2010) coincided for countries for which data is available in both databases. In this way, we can be relatively sure that the same definitions of immigration are used in both databases and that the consistency of our dataset is not compromised by combining them. E.g., Eurostat (2010) data could not be used with regard to Italy and Portugal, as data available in both databases deviates considerably, so we decided to remain with what is provided by OECD (2010). One disadvantage of adding the Eurostat (2010) data is that the series starts in 1998, so that previous years remain unchanged. For Switzerland, this results in a sharp increase in inflows from 12,656 in 1997 to 32,179 in 1998 simply due to the increased number of source countries in the latter year. As a consequence, the data probably does not reflect the actual development of immigration flows at this period, but this drawback is likely to be compensated by the additional observations for all following years. The picture is friendlier with regard to the Scandinavian countries and the Netherlands, whereby observations of inflows into the latter were also supplemented with data from Eurostat (2010). However, the magnitude of immigration is rather small in these countries. Fortunately, the data is predominantly complete for France, Spain and Germany, as these three countries together represent about two-thirds of the migratory flows to Europe in our sample. Data from Australia, Canada and the United States covers inflows from almost all source countries for all or nearly all time periods, reflecting the long tradition of immigration to these countries and the attention that 8

10 is paid to immigration by the authorities. With 12 years, 161 countries of origin and 19 countries of destination, we would obtain a maximum of 42,826 observations. However, 17,331 of the observations remain missing. But despite this and other undeniable weak points, we can claim to have obtained a comprehensive dataset of immigration to OECD countries by combining OECD (2010) and Eurostat (2010) where it was possible. Data for most of the economic and demographic variables is obtained from the World Bank s World Development Indicators (WDI) (The World Bank 2010) database, which is available online. Since the GDP per capita does not take into account the income distribution (e.g. migrants from rural areas might earn much less in their home countries) selecting GDP per capita would underestimate the income differential. However, data on possible alternatives, as e.g. workers or farmers incomes, is not sufficiently available from developing countries, so that we have to remain with GDP per capita. For some variables, e.g. unemployment in the source countries, data is frequently missing. The situation is probably worst regarding the Gini coefficient and related measures of poverty and inequality, as a larger number of time periods is only available for some countries of Southern America and Middle and Eastern Europe (MEE). Other gravity variables, in particular distances, contiguity, common language and colony are obtained from CEPII. 4.2 Estimates of the static model In our empirical model, we have specified source and destination country fixed effects and country-pair effects respectively. Since we cannot always assume that these effects have zero mean and are uncorrelated with the regressors, we rule out the use of a random effects model. Instead, we assume that these effects are not random, but fixed and correlated with the explanatory variables. Possible reasoning behind this assumption are e.g. that each destination country has its own unobservable, time-constant mentality towards immigration that affects 9

11 actual immigration flows, or that there exist specific relations between some source and destination countries that we do not control for with our regressors. The country or country-pair unobserved heterogeneity could be eliminated by applying the FE estimator, as time-demeaning wipes out all time-invariant effects. However, this would mean that we also lose all timeinvariant variables from our model and thereby presumably important information. In our case, we would not be able to investigate the importance of distance between the source and destination countries or of identical languages, which can be considered as a high price to pay. Fortunately, we can avoid losing the time-invariant regressors and still obtain consistent estimates of the parameters by applying the so-called Least Squares Dummy Variable (LSDV) estimator with exporter and importer effects. The idea is not to remove the FE, but to estimate a separate intercept for each group in the panel. Thus, dummy variables for each source and destination country are generated and are included in an OLS regression. With regard to our sample, applying LSDV implies including 180 dummies for country FE and 3,059 dummies for countrypair FE respectively, which is computationally extensive, but still workable. One deviation from our empirical model we have to make is that we omit the unemployment in the source countries as an explanatory variable from our main regressions. This is because more than 50 percent of the observations for this variable are missing in our dataset, which would imply a huge loss of information if we included it in the regressions. Table 1 contains summary statistics of the data. Table 2 shows the results obtained from estimating specifications (1) and (2) using several methodologies and using migration flows as dependent variable. Column 1 contains the estimates from a LSDV regression with only time dummies, column 2 adding source and destination country dummies, while column 3 adds country pair dummy variables and drops the bilateral 10

12 time invariant variables. Finally, column 4 estimates specification (2) with multilateral resistance dummies for each destination-and-time period. Table 3 also shows the results obtained from estimating specifications (1) and (2) but the dependent variable used is the stock of migrants in the destination country, whereas in Table 4 the dependent variable is the migration rate. As regard to the climate variables, we interpret the results obtained in our preferred specification (2) that controls for all the unobservable heterogeneity that is specific to the host countries and varies over time and for bilateral unobservable heterogeneity. Considering the emigration rate as dependent variable, the estimated elasticity for the weighted temperature variable is (Table 4, Model 4), indicating that an increase in average temperature of 10 percent increases migration by 2.6 percent. The estimated elasticity for the weighted precipitation variable is (Table 4, Model 4), indicating that a decrease in average precipitation of 10 percent increases migration by 0.5 percent. To translate these estimates into numbers, we use the average temperature (19 degrees Celsius) and average precipitation (11 mm) in our sample and the coefficient estimated for the migration stock. Using the results in Table 3 column 4, we find that, in static terms, the average increase in migration stock is approximately 761 migrants for every increase in 1 degree Celsius in the origin country. This average is calculated as Migrants _ stock wtemp Emig _ rate Migrants _ stock lwtemp * lwtemp * wtemp Migrants _ stock wtemp wtemp * With respect to the average effect of variations in precipitation, using the estimates in Table 2, column 4, in which the dependent is migration flows and using a similar calculation as before we find that, in static terms, a decrease in precipitation of 1mm per year will increase migration flows by approximately 7 migrants from each origin to each destination. 11

13 4.3 Dynamic model results The results obtained from estimating model 3 in with dynamics and network effects are modeled are presented in Table 5. The first part of Table 5 uses migration flows as dependent variable, whereas the second presents results using emigration rates. In both cases we find positive network effects: an increase in past stock of migrants of 10 percent has a positive and significant effect on the present inflow, with an increase of around 1 percent. The adjustment mechanism is also positive, the actual inflow increases with the past inflow of migrants. We find a high degree of persistence with estimated coefficients for the lagged dependent variable between 0.41 and Since the time dimension of our data set is only 12 years, we are aware of the fact that the estimated coefficients are biases and the true effect will lie between the OLS and the two way FE estimates. As regards the climate variables, only the coefficient of average precipitation is statistically significant, indicated that a decrease in precipitation of 10 percent will increase migration flows by almost 1 percent. This should be interpreted as a short run effect, the long run effect is much larger than the short run effect lying between 2 to 6 percent will increase migration flows associate to the same decrease in precipitation. The average temperature variable is not statistically significant in the dynamic specification, perhaps do to the short time period used in the estimations. Robustness: zero values, gmm. 4.4 Predictions Based on Marchiori (2012) method. 12

14 5. Conclusions This paper is the first to use international migration data and climate variables in a multi-country setting to investigate to what extent migration can be explained with climate variations. A gravity model of migration augmented with average temperature and precipitation in the origin country is estimated including the usual control variables. Several panel data estimators are applied, including dynamic panel data models. Our main findings are twofold. First, higher temperatures increase migration for all countries in a static setting, whereas with the incorporation of dynamic effects and network effects in the model this effect vanishes. Second, lower precipitation levels also increase migration both in the short and in the long term and the effect remains when dynamics and network effects are modeled. Both effects are significant in statistical and in economic terms. In particular, according to the results from the static model a 1 C rise in average temperature in a given year increases bilateral migration flows by about four percent per destination on average, which is equivalent to around 761 migrants. Also in static terms, a decrease in precipitation of 1mm per year will increase migration flows by approximately 7 migrants from each origin to each destination. Using results from the dynamic model the effects are more pronounced, specially the long run effect indicate that a decrease in precipitation of 1mm per year will be associated to an increase in migration flows by approximately migrants from each origin to each destination per year. 13

15 FIGURES Figure 1. Total immigrant flow by destination country Australia Austria Belgium Canada Denmark Finland France Germany Italy Japan tinflowsm Korea Netherlands Norway Portugal Spain Sweden Switzerland United Kingdom United States Graphs by destination year Note: Reported figures are in thousand people. 14

16 TABLES Table 1. Summary Statistics Variables in ln Obs Mean Std. Dev. Min Max ln_inflows ln_stocks ln_emig_rate ln_wtemperature_origin ln_wprecipitation_origin ln_gdp_destination ln_gdp_origin ln_pop_origin Demographic pressure Unemployment origin unemployment destination Trade_to_gdp ln_distance Contiguity Same_continent Language Colony EU membership Stocks E+07 Inflows Emigration Rate Wtem_Or Wpre_Or

17 Table 2. Main determinants of bilateral migration flows. Static model dep var : ln_migration_flow LSDV with time dummies LSDV with time and country dummies LSDV with time and pair dummies LSDV with host country-and-time and pair dummies Independent variables: M1 M2 M3 M4 b/se b/se b/se ln_wtemperature_origin se (0.082) (0.206) (0.146) (0.132) ln_wprecipitation_origin ** * ** se (0.049) (0.030) (0.025) (0.022) ln_gdp_destination 3.154*** 1.851*** 1.568*** se (0.525) (0.274) (0.253) ln_gdp_origin 2.299*** 1.759*** 1.480*** 1.505*** se (0.421) (0.575) (0.481) (0.443) ln_gdp_or_sq *** *** *** *** se (0.027) (0.035) (0.029) (0.027) Ln_population_origin 0.678*** ** *** se (0.029) (0.394) (0.323) (0.297) demographic_preasure 0.109*** ** se (0.012) (0.014) (0.012) (0.011) unemployment_destination 0.126*** *** *** se (0.019) (0.011) (0.010) trade_to_gdp ratio ** 0.002** 0.001** 0.001** se (0.001) (0.001) (0.001) (0.001) ln_distance se (0.045) (0.034) contiguity se (0.403) (0.214) same_continent 1.368*** 1.527*** se (0.140) (0.132) common language 1.354*** 0.982*** se (0.135) (0.124) Colonial relationship 2.168*** 2.133*** se (0.170) (0.173) European Union Membership *** se (0.101) (0.044) (0.032) (0.057) R-squared N ll rmse Note:***,**,* denote significance levels at the one, five and ten percent, respectively. 16

18 Table 3. Main determinants of bilateral migration stocks. Static model dep var: ln_migration_stock LSDV with time dummies 17 LSDV with time and country dummies LSDV with time and pair dummies Independent variables: M1 M2 M3 M4 LSDV with host country-andtime and pair dummies b/se b/se b/se ln_wtemperature_origin ** 0.389** 0.395** se (0.133) (0.259) (0.172) (0.161) ln_wprecipitation_origin se (0.079) (0.038) (0.023) (0.021) ln_gdp_destination 2.422*** 1.878*** 1.384*** se (0.529) (0.298) (0.234) ln_gdp_origin 3.276*** *** 1.829*** se (0.644) (0.703) (0.507) (0.492) ln_gdp_or_sq *** *** *** se (0.040) (0.043) (0.031) (0.030) Ln_population_origin 0.825*** *** *** *** se (0.045) (0.517) (0.401) (0.388) demographic_preasure 0.087*** 0.036** * se (0.019) (0.016) (0.011) (0.011) unemployment_destination *** *** se (0.021) (0.011) (0.009) trade_to_gdp ratio se (0.002) (0.001) (0.001) (0.001) ln_distance 0.215*** 0.092** se (0.070) (0.046) contiguity 1.153** 0.408* se (0.541) (0.247) same_continent 1.280*** 1.524*** se (0.213) (0.195) common language 2.432*** 0.844*** se (0.250) (0.219) Colonial relationship 2.207*** 2.158*** se (0.315) (0.298) European Union Membership *** 0.106*** se (0.164) (0.056) (0.031) (0.066) R-squared N ll rmse Note:***,**,* denote significance levels at the one, five and ten percent, respectively.

19 Table 4. Main determinants of bilateral migration rate. Static model dep var: ln_migration_rate LSDV with time dummies LSDV with time and country dummies LSDV with time and pair dummies Independent variables: M1 M2 M3 M4 LSDV with host country-andtime and pair dummies b/se b/se b/se b/se ln_wtemperature_origin * 0.265** se (0.085) (0.210) (0.148) (0.134) ln_wprecipitation_origin ** ** ** se (0.050) (0.030) (0.024) (0.022) ln_gdp_destination 3.523*** 1.821*** 1.535*** se (0.547) (0.279) (0.258) ln_gdp_origin 1.704*** 0.940* se (0.430) (0.524) (0.443) (0.407) ln_gdp_or_sq *** * se (0.027) (0.030) (0.025) (0.023) demographic_preasure 0.068*** se (0.012) (0.014) (0.012) (0.011) unemployment_destination 0.136*** *** *** se (0.020) (0.011) (0.010) trade_to_gdp ratio 0.004*** 0.002*** 0.002*** 0.002*** se (0.001) (0.001) (0.001) (0.001) ln_distance se (0.046) (0.034) contiguity se (0.393) (0.214) same_continent 1.432*** 1.528*** se (0.143) (0.132) common language 1.522*** 0.983*** se (0.141) (0.124) colonial relationship 2.152*** 2.138*** se (0.178) (0.173) EU membership *** se (0.103) (0.044) (0.032) (0.058) R-squared N ll rmse Note:***,**,* denote significance levels at the one, five and ten percent, respectively. 18

20 Table 5: Dynamic models for inflows and migration rate Dep var ln_inflows ln_emig_rate ols, time dummies time and i,j, fe time and pair fe ols, time dummies time and i,j, fe time and pair fe Indep. Variables: b/se b/se b/se b/se b/se b/se L.ln_inflows 0.873*** 0.740*** 0.409*** (0.010) (0.015) (0.023) L.ln_emig_rate 0.927*** 0.740*** 0.409*** (0.006) (0.015) (0.023) L.ln_stocks 0.078*** 0.172*** 0.148*** 0.026*** 0.172*** 0.147*** (0.009) (0.014) (0.034) (0.005) (0.014) (0.034) ln_wtem_or * (0.015) (0.181) (0.169) (0.013) (0.181) (0.169) ln_wpre_or *** *** *** *** (0.008) (0.033) (0.030) (0.007) (0.033) (0.030) ln_gdp_destination 0.209*** ** 0.244*** ** (0.068) (0.184) (0.223) (0.066) (0.184) (0.223) ln_gdp_origin (0.102) (0.369) (0.506) (0.094) (0.369) (0.506) ln_gdp_or_sq ** ** (0.006) (0.023) (0.030) (0.006) (0.023) (0.030) ln_pop_or 0.047*** 0.559** (0.006) (0.276) (0.345) (0.275) (0.346) demographic_preasure 0.004* (0.002) (0.009) (0.013) (0.002) (0.009) (0.013) unemployment_dest 0.037*** *** 0.036*** *** (0.004) (0.007) (0.008) (0.004) (0.007) (0.008) trade_to_gdp ** 0.001*** ** (0.000) (0.001) (0.001) (0.000) (0.001) (0.001) same_continent 0.182*** 0.292*** 0.156*** 0.292*** (0.022) (0.034) (0.019) (0.034) Language 0.070** 0.187*** 0.114*** 0.187*** (0.032) (0.043) (0.029) (0.043) Colony 0.227*** 0.160*** 0.193*** 0.161*** (0.042) (0.057) (0.037) (0.057) Eu *** *** (0.017) (0.025) (0.029) (0.016) (0.025) (0.029) R-squared N Ll Rmse Note:***,**,* denote significance levels at the one, five and ten percent, respectively. 19

21 Table 6. Dynamic model estimated in first differences and with IV Instrumental Variables Instrumental Variables Dep variables: Inflows Ln_emig_rate Indep. Var: b/se b/se LD.ln_inflows * (0.358) LD.ln_emig_rate * (0.362) LD.ln_stocks (0.271) (0.274) LD.ln_wtem_or (0.180) (0.181) LD.ln_wpre_or *** *** (0.029) (0.029) D.ln_gdp_destin (0.303) (0.304) D.ln_gdp_origin (1.174) (1.180) D.ln_gdp_or_sq (0.072) (0.072) D.ln_pop_or (0.984) (0.994) D.demographic_de (0.040) (0.040) D.unemploymen or *** *** (0.026) (0.027) D.trade_to_gdp (0.001) (0.001) D.eu 0.246*** 0.246*** (0.046) (0.046) cons (0.039) (0.039) R-squared N ll rmse j jp Note:***,**,* denote significance levels at the one, five and ten percent, respectively. 20

22 Table 7. Heckman 2-steps results DEP VAR Inflows ln_inflows inflows b/se b/se ln_wtem_or 0.313*** ln_wpre_or ** *** ln_gdp_destina~n 1.151*** 0.617*** ln_gdp_origin 1.335*** 1.506*** ln_gdp_or_sq *** *** demographic_pr~e 0.018*** 0.071*** unemployment_d~n *** ** trade_to_gdp 0.002*** *** 0 0 same_continent 1.467*** 0.826*** language *** 0.467*** colony 2.291*** 0.742*** stability *** year R-squared r2_w N ll Note:***,**,* denote significance levels at the one, five and ten percent, respectively. 21

23 References Afifi, T.; Warner, K. (2008): The Impact of Environmental Degradation on Migration Flows across Countries. Working Paper No.5/2008, UNU-EHS, Bonn. Baldwin and Taglioni, 2006 Black, Richard, W. Neil Adger, Nigel W. Arnell, Stefan Dercon, Andrew Geddes, David Thomas The effect of environmental change on human migration, Global Environmental Change 21, Supplement 1, December, S3-S11. CEPII (2011): Centre D Etudes Prospectives Et D Informations Internationales: < Dell, M. Jones, B. F. and Olken, B. O. (2008) Climate Change and Economic Growth: Evidence from the Last Half Century, NBER Working Paper Drabo, A. and Mbaye, L. (2011), Climate Change, Natural Disasters and Migration: An Empirical Analysis in Developing Countries, IZA Discussion Paper Available at SSRN: Dunlevy, J. A. (1993) "Migrant Stock vs. Lagged Migrant Flow as a Determinant of Migrant Settlement: A Comment," Journal of Population Economics, Springer, vol. 6(2), Hatton, T. J. and Williamson, J. G. (2003) "Demographic and Economic Pressure on Emigration out of Africa," Scandinavian Journal of Economics, Wiley Blackwell, vol. 105(3), pages , 09. Karemera, D., et al (2000) A gravity model analysis of international migration to North America, Applied Economics 32 (13), Lewer, J. and H. Van den Berg A Gravity Model of Immigration. Economics letters, 99(1): Lilleor, H. B. and Van den Broeck, K. (2011) Economic drivers of migration and climate change in LDCs, Global Environmental Change 21 (1), S70-S81. Mayda, A International Migration: a Panel Data Analysis of the Determinants of Bilateral Flows. Journal of Population Economics, 23: Marchiori, Luca, Jean-Francois Maystadt, Ingmar Schumacher, (2012). The impact of climate variations on migration in sub-saharan Africa, Journal of Environmental Economics and Management 63 (3), Ruyssen, Ilse, Everaert, Gerdie and Rayp, Glenn (2011). Determinants of International Migration to the OECD in a Three-Way Dynamic Panel Framework. SHERPPA, Ghent University. Mimeo. Warner, K.; Stal, M.; Dun, O.; Afifi, T. (2009). Researching environmental change and migration: Evaluation of EACH-FOR methodology and application in 23 case studies worldwide. In: Laczko, F.; Aghazarm, C. (Eds.): Migration, Environment and Climate Change: Assessing the Evidence. IOM, Geneva, pp Warin, T. and Svaton, P. (2008) European Migration: Welfare Migration or Economic Migration? Global Economic Journal 8 (3),

24 APPENDIX Table A.1. List of countries origin destination Afghanistan Egypt Lesotho Senegal Australia Albania El Salvador Liberia Sierra Leone Austria Algeria Equatorial Guinea Libya Slovakia Belgium Angola Eritrea Lithuania Slovenia Canada Argentina Estonia Luxembourg Solomon Islands Denmark Armenia Ethiopia Macedonia Somalia Finland Australia Fiji Madagascar South Africa France Austria Finland Malawi Spain Germany Azerbaijan France Malaysia Sri Lanka Italy Bahamas French Guiana Mali Sudan Japan Bangladesh French Polynesia Mauritania Suriname Korea Belarus Gabon Mauritius Swaziland Netherlands Belgium Gambia Mexico Sweden Norway Belize Georgia Moldova Switzerland Portugal Benin Germany Mongolia Syria Spain Bhutan Ghana Morocco Taiwan Sweden Bolivia Greece Mozambique Tajikistan Switzerland Bosnia and Herzegovina Greenland Myanmar Tanzania United Kingdom Botswana Guadeloupe Namibia Thailand United States Brazil Guatemala Nepal Timor-Leste Brunei Darussalam Guinea Netherlands Togo Bulgaria Guinea-Bissau New Caledonia Trinidad and Tobago Burkina Faso Guyana New Zealand Tunisia Burundi Haiti Nicaragua Turkey Cambodia Honduras Niger Cameroon Hungary Nigeria Canada Iceland Norway Cape Verde India Oman Central African Republic Indonesia Pakistan Chad Iran Panama Chile Iraq Papua New Guinea China Ireland Paraguay Colombia Israel Peru Comoros Italy Philippines Congo (Brazzaville) Jamaica Poland Congo (Kinshasa) Japan Portugal Costa Rica Jordan Puerto Rico Croatia Kazakhstan Qatar Cuba Kenya Reunion 23

25 Cyprus Korea, North Romania Czech Republic Korea, South Russian Federation Cote d'ivoire Kuwait Rwanda Denmark Kyrgyzstan Saint Vincent and the Grenadines Djibouti Laos Samoa Dominican Republic Latvia Sao Tome and Principe Ecuador Lebanon Saudi Arabia 24

26 Table A.2. List of variables and sources Variable Definition Source Wtemperature_origin Population weighted average annual temperature in degrees C. Dell et al. (2008). Constant 1990 population weights Wprecipitation_origin Population weighted average annual precipitation in mm. Constant Dell et al. (2008) population weights Immigrant inflow as a Percentage of host Country population Inflow of foreign population into the host country as a percentage of host country population. Organization for Economic Co-operation and Development (OECD) OECD.StatExtracts. Paris: OECD Publishing. Stock of immigrant Population Stock of foreign population by nationality in the host country as a percentage of host country population. Organization for Economic Co-operation and Development (OECD) OECD.StatExtracts. Paris: OECD GDP per capita origin PPP adjusted GDP per capita in origin at current US$ World Bank World Development Indicators Database GDP per capita PPP adjusted GDP per capita in host country, current US$ World Bank World Development Indicators Database destination Population origin Population in the origin country(number of inhabitants) World Bank World Development Indicators Database Demographic pressure Unemployment rate Origin Unemployment rate Destination Percentage or young population as a share of working age population Unemployment rate in the host country, total (percent of total labor force). Unemployment rate in the origin country, total (percent of total labor force). World Bank World Development Indicators Database The World Bank: Washington. World Bank World Development Indicators. Vol The World Bank: Washington. World Bank World Development Indicators. Vol The World Bank: Washington. Same Continent Dummy variable equals 1 if two countries are in the same continent Centre D'Etudes Prospectives et D'Informations Internationales (CEPII) CEPII Distance Database. Contiguity Dummy variable equals 1 if two countries are contiguous. CEPII Distance Database. Common language Dummy variable equals 1 if two countries share a official language. CEPII Distance Database. Colony Distance European union (EU) Dummy variable equals 1 if two countries ever had a colonial link. Geodesic distance calculated by the great circle formula using latitude and longitude of the most Dummy variable equals 1 if two countries are members of the European Union CEPII Distance Database. CEPII Distance Database. World Trade Organization